Benchmark RadarAI BENCHMARK PROFILE
PRMU
PRMU evaluates corpus-free multimodal unlearning of person-related knowledge in MLLMs, using textual and visual probes including adversarial evaluation and locality analysis.
- Released
- 2026-08-11
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Addresses realistic deletion scenarios where original corpora are unavailable, providing a way to measure forgetting-locality trade-offs and vulnerability to knowledge reactivation, which is valuable for safe deployment.
Motivation
Multimodal large language models (MLLMs) have demonstrated remarkable capabilities in storing and recalling rich person-related knowledge, raising increasing concerns about reliable knowledge removal.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.